'return' or 'exit' in code talk
A neuronOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in layerOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. 8 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featuredA neuron picked for the full treatment: steering outputs and a written explanation. About 3,000 of the 86,000 neurons are featured., so it has steering outputs.
What this neuron does
This neuron fires on words about returning a result or leaving a function, such as "return" in a coding explanation or "exit" for leaving a loop or recursion. Turning it up led to talk about how to handle outcomes, while turning it down led to a rambling, repetitive answer.
Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens a prefilled issue on GitHub).
Fires most on
The text windows where this neuron's activationThe number a neuron produces for a token. Large positive means it fired hard; negative means it pushed the other way. was highest. Amber marks positive activation; the outlined tokenA piece of text the model reads at once: often a word, sometimes part of a word or a punctuation mark. is the peak. Hover a token for its strength.
-
True, obj)". If it failed, you might return "(False, reason)". Often, you can use exceptions to handle this situation.⏎Sometimes you might want to return objects of different types based on arguments to the function. For instance, did the
-
looks like this:⏎following as appropriate:⏎This is a bool to indicate success.⏎If unsuccessful, this is a Pylons response object that your⏎controller can return. It will contain an appropriate error⏎message for the user.⏎Upon success,
-
This increases durability and ensures that you can continue using this set again and again.⏎How will you receive the resource?⏎After your payment has processed you will be redirected to a download page where you can immediately download the file and save it to your computer
-
having the x if you set total to 0 at the start because total will never increase, and that's even if you ignore that there is no way to exit the recursion.⏎Also, because total is a local variable and you
-
don't know what happened to it, but I probably never completed it anyways.⏎And that recursive function looks really....recursive. I don't see any code to exit out of the loop, IE it will continue to call it self over and over
-
)".⏎Sometimes you might want to return objects of different types based on whether an operation succeeded or not. For instance, if the operation was successful, you might return "(
SteeringAdding a push along a neuron while the model writes, to see how its output changes. Turning it up or down.
Push this neuron up or down while the model writes, and watch what changes. The outputs were generated ahead of time, so nothing runs in your browser.
Here is something I have been thinking about lately: 1. A. Monday-Friday: Monday-Sunday (Spa Day) 2. Wednesday-Thursday (Spa Day) 3. Friday (Special Day) 4. Special Event 5. Last Week's Event
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
expression were noted in transplant recipients. These studies suggest a specific strategy for LAD gene therapy that may be effective and safe.<|endoftext|>The resource has been added to your collection⏎This interactive activity adapted from NOVA Online features animations that illustrate the components
-
words by clicking on the first and last letters, in any order. If the selected letters match an item in the list, the letters are highlighted and the item is crossed out. Spaces, capitalization and diàçritîcš are
-
. It is an extension of Bullseye matrix problem.⏎Who has power to do everything in this world?⏎Close MATLAB with keyboard⏎Find the treasures in MATLAB Central and discover how the community can help you!⏎Choose a web site to get translated
-
, to the assets of the company in the event of its dissolution. Dividends are usually set at a fixed annual rate that must be paid before dividends are distributed to the common stockholders. Preferred stock may be participating, meaning that its holders
Connections
Show on mapWhich neuronsOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in earlier layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.
Fed by earlier layers
- L5:688 excites
- L7:390 suppresses
- L7:906 suppresses
- L6:2618 excites
- L6:652 suppresses
- L7:1840 weakly suppresses
- L7:2065 weakly suppresses
- L7:648 weakly suppresses
- L6:693 weakly suppresses
- L2:2662 weakly excites
- L6:2115 weakly excites
- L2:558 weakly suppresses
- L4:1413 weakly excites
- L6:1515 weakly suppresses
- L7:2383 weakly excites
- L7:274 weakly excites
- L4:824 weakly suppresses
- L6:971 weakly excites
- L7:641 weakly excites
- L6:1488 weakly excites
- L4:14 weakly suppresses
- L1:2944 weakly excites
- L7:1048 weakly excites
- L7:1146 weakly excites
Feeds later layers
- L17:2151 excites
- L12:492 excites
- L11:1030 excites
- L10:1591 suppresses
- L12:340 suppresses
- L11:1350 excites
- L13:1953 suppresses
- L10:1238 excites
- L18:1276 excites
- L11:679 weakly excites
- L11:1343 weakly excites
- L10:2726 weakly suppresses
- L10:1691 weakly suppresses
- L11:1702 weakly suppresses
- L11:1029 weakly suppresses
- L11:843 weakly excites
- L15:1863 weakly excites
- L10:2602 weakly excites
- L12:369 weakly excites
- L11:1432 weakly excites
- L11:2216 weakly excites
- L21:2978 weakly suppresses
- L10:1554 weakly excites
- L21:2970 weakly suppresses
Activation histogramA chart of how often a neuron takes each activation value.
How often this neuron fires at each strength. Almost always it does close to nothing (the tall bars in the middle); the snippets above are the rare moments it fired hard, amber for positive and blue for negative.
Logit lensReading which output tokens a neuron pushes toward or away from, straight off its output weights, before the later layers have their say.
The words this neuron nudges the model toward (amber) or away from (blue) when it fires. A hint, not a label: many neurons work through later layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. instead.
Promotes
- 婴
- úb
- abound
- ifier
- WithValue
- 的地图
- subordinate
- 绪
- ui
- 的心情
Suppresses
- rés
- achel
- ��
- cip
- artment
- amines
- oxetine
- eni
- exem
- 物理学